projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Drop 'size' argument from bio_endio and bi_end_io
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / raid1.c
1 /*
2 * raid1.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
5 *
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
7 *
8 * RAID-1 management functions.
9 *
10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
11 *
12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
14 *
15 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
16 * bitmapped intelligence in resync:
17 *
18 * - bitmap marked during normal i/o
19 * - bitmap used to skip nondirty blocks during sync
20 *
21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
22 * - persistent bitmap code
23 *
24 * This program is free software; you can redistribute it and/or modify
25 * it under the terms of the GNU General Public License as published by
26 * the Free Software Foundation; either version 2, or (at your option)
27 * any later version.
28 *
29 * You should have received a copy of the GNU General Public License
30 * (for example /usr/src/linux/COPYING); if not, write to the Free
31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
32 */
33
34 #include "dm-bio-list.h"
35 #include <linux/raid/raid1.h>
36 #include <linux/raid/bitmap.h>
37
38 #define DEBUG 0
39 #if DEBUG
40 #define PRINTK(x...) printk(x)
41 #else
42 #define PRINTK(x...)
43 #endif
44
45 /*
46 * Number of guaranteed r1bios in case of extreme VM load:
47 */
48 #define NR_RAID1_BIOS 256
49
50
51 static void unplug_slaves(mddev_t *mddev);
52
53 static void allow_barrier(conf_t *conf);
54 static void lower_barrier(conf_t *conf);
55
56 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
57 {
58 struct pool_info *pi = data;
59 r1bio_t *r1_bio;
60 int size = offsetof(r1bio_t, bios[pi->raid_disks]);
61
62 /* allocate a r1bio with room for raid_disks entries in the bios array */
63 r1_bio = kzalloc(size, gfp_flags);
64 if (!r1_bio)
65 unplug_slaves(pi->mddev);
66
67 return r1_bio;
68 }
69
70 static void r1bio_pool_free(void *r1_bio, void *data)
71 {
72 kfree(r1_bio);
73 }
74
75 #define RESYNC_BLOCK_SIZE (64*1024)
76 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
77 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
78 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
79 #define RESYNC_WINDOW (2048*1024)
80
81 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
82 {
83 struct pool_info *pi = data;
84 struct page *page;
85 r1bio_t *r1_bio;
86 struct bio *bio;
87 int i, j;
88
89 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
90 if (!r1_bio) {
91 unplug_slaves(pi->mddev);
92 return NULL;
93 }
94
95 /*
96 * Allocate bios : 1 for reading, n-1 for writing
97 */
98 for (j = pi->raid_disks ; j-- ; ) {
99 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
100 if (!bio)
101 goto out_free_bio;
102 r1_bio->bios[j] = bio;
103 }
104 /*
105 * Allocate RESYNC_PAGES data pages and attach them to
106 * the first bio.
107 * If this is a user-requested check/repair, allocate
108 * RESYNC_PAGES for each bio.
109 */
110 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
111 j = pi->raid_disks;
112 else
113 j = 1;
114 while(j--) {
115 bio = r1_bio->bios[j];
116 for (i = 0; i < RESYNC_PAGES; i++) {
117 page = alloc_page(gfp_flags);
118 if (unlikely(!page))
119 goto out_free_pages;
120
121 bio->bi_io_vec[i].bv_page = page;
122 }
123 }
124 /* If not user-requests, copy the page pointers to all bios */
125 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
126 for (i=0; i<RESYNC_PAGES ; i++)
127 for (j=1; j<pi->raid_disks; j++)
128 r1_bio->bios[j]->bi_io_vec[i].bv_page =
129 r1_bio->bios[0]->bi_io_vec[i].bv_page;
130 }
131
132 r1_bio->master_bio = NULL;
133
134 return r1_bio;
135
136 out_free_pages:
137 for (i=0; i < RESYNC_PAGES ; i++)
138 for (j=0 ; j < pi->raid_disks; j++)
139 safe_put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page);
140 j = -1;
141 out_free_bio:
142 while ( ++j < pi->raid_disks )
143 bio_put(r1_bio->bios[j]);
144 r1bio_pool_free(r1_bio, data);
145 return NULL;
146 }
147
148 static void r1buf_pool_free(void *__r1_bio, void *data)
149 {
150 struct pool_info *pi = data;
151 int i,j;
152 r1bio_t *r1bio = __r1_bio;
153
154 for (i = 0; i < RESYNC_PAGES; i++)
155 for (j = pi->raid_disks; j-- ;) {
156 if (j == 0 ||
157 r1bio->bios[j]->bi_io_vec[i].bv_page !=
158 r1bio->bios[0]->bi_io_vec[i].bv_page)
159 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
160 }
161 for (i=0 ; i < pi->raid_disks; i++)
162 bio_put(r1bio->bios[i]);
163
164 r1bio_pool_free(r1bio, data);
165 }
166
167 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
168 {
169 int i;
170
171 for (i = 0; i < conf->raid_disks; i++) {
172 struct bio **bio = r1_bio->bios + i;
173 if (*bio && *bio != IO_BLOCKED)
174 bio_put(*bio);
175 *bio = NULL;
176 }
177 }
178
179 static void free_r1bio(r1bio_t *r1_bio)
180 {
181 conf_t *conf = mddev_to_conf(r1_bio->mddev);
182
183 /*
184 * Wake up any possible resync thread that waits for the device
185 * to go idle.
186 */
187 allow_barrier(conf);
188
189 put_all_bios(conf, r1_bio);
190 mempool_free(r1_bio, conf->r1bio_pool);
191 }
192
193 static void put_buf(r1bio_t *r1_bio)
194 {
195 conf_t *conf = mddev_to_conf(r1_bio->mddev);
196 int i;
197
198 for (i=0; i<conf->raid_disks; i++) {
199 struct bio *bio = r1_bio->bios[i];
200 if (bio->bi_end_io)
201 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
202 }
203
204 mempool_free(r1_bio, conf->r1buf_pool);
205
206 lower_barrier(conf);
207 }
208
209 static void reschedule_retry(r1bio_t *r1_bio)
210 {
211 unsigned long flags;
212 mddev_t *mddev = r1_bio->mddev;
213 conf_t *conf = mddev_to_conf(mddev);
214
215 spin_lock_irqsave(&conf->device_lock, flags);
216 list_add(&r1_bio->retry_list, &conf->retry_list);
217 conf->nr_queued ++;
218 spin_unlock_irqrestore(&conf->device_lock, flags);
219
220 wake_up(&conf->wait_barrier);
221 md_wakeup_thread(mddev->thread);
222 }
223
224 /*
225 * raid_end_bio_io() is called when we have finished servicing a mirrored
226 * operation and are ready to return a success/failure code to the buffer
227 * cache layer.
228 */
229 static void raid_end_bio_io(r1bio_t *r1_bio)
230 {
231 struct bio *bio = r1_bio->master_bio;
232
233 /* if nobody has done the final endio yet, do it now */
234 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
235 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
236 (bio_data_dir(bio) == WRITE) ? "write" : "read",
237 (unsigned long long) bio->bi_sector,
238 (unsigned long long) bio->bi_sector +
239 (bio->bi_size >> 9) - 1);
240
241 bio_endio(bio,
242 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
243 }
244 free_r1bio(r1_bio);
245 }
246
247 /*
248 * Update disk head position estimator based on IRQ completion info.
249 */
250 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
251 {
252 conf_t *conf = mddev_to_conf(r1_bio->mddev);
253
254 conf->mirrors[disk].head_position =
255 r1_bio->sector + (r1_bio->sectors);
256 }
257
258 static void raid1_end_read_request(struct bio *bio, int error)
259 {
260 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
261 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
262 int mirror;
263 conf_t *conf = mddev_to_conf(r1_bio->mddev);
264
265 mirror = r1_bio->read_disk;
266 /*
267 * this branch is our 'one mirror IO has finished' event handler:
268 */
269 update_head_pos(mirror, r1_bio);
270
271 if (uptodate)
272 set_bit(R1BIO_Uptodate, &r1_bio->state);
273 else {
274 /* If all other devices have failed, we want to return
275 * the error upwards rather than fail the last device.
276 * Here we redefine "uptodate" to mean "Don't want to retry"
277 */
278 unsigned long flags;
279 spin_lock_irqsave(&conf->device_lock, flags);
280 if (r1_bio->mddev->degraded == conf->raid_disks ||
281 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
282 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)))
283 uptodate = 1;
284 spin_unlock_irqrestore(&conf->device_lock, flags);
285 }
286
287 if (uptodate)
288 raid_end_bio_io(r1_bio);
289 else {
290 /*
291 * oops, read error:
292 */
293 char b[BDEVNAME_SIZE];
294 if (printk_ratelimit())
295 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
296 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
297 reschedule_retry(r1_bio);
298 }
299
300 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
301 }
302
303 static void raid1_end_write_request(struct bio *bio, int error)
304 {
305 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
306 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
307 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
308 conf_t *conf = mddev_to_conf(r1_bio->mddev);
309 struct bio *to_put = NULL;
310
311
312 for (mirror = 0; mirror < conf->raid_disks; mirror++)
313 if (r1_bio->bios[mirror] == bio)
314 break;
315
316 if (error == -EOPNOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
317 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
318 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
319 r1_bio->mddev->barriers_work = 0;
320 /* Don't rdev_dec_pending in this branch - keep it for the retry */
321 } else {
322 /*
323 * this branch is our 'one mirror IO has finished' event handler:
324 */
325 r1_bio->bios[mirror] = NULL;
326 to_put = bio;
327 if (!uptodate) {
328 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
329 /* an I/O failed, we can't clear the bitmap */
330 set_bit(R1BIO_Degraded, &r1_bio->state);
331 } else
332 /*
333 * Set R1BIO_Uptodate in our master bio, so that
334 * we will return a good error code for to the higher
335 * levels even if IO on some other mirrored buffer fails.
336 *
337 * The 'master' represents the composite IO operation to
338 * user-side. So if something waits for IO, then it will
339 * wait for the 'master' bio.
340 */
341 set_bit(R1BIO_Uptodate, &r1_bio->state);
342
343 update_head_pos(mirror, r1_bio);
344
345 if (behind) {
346 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
347 atomic_dec(&r1_bio->behind_remaining);
348
349 /* In behind mode, we ACK the master bio once the I/O has safely
350 * reached all non-writemostly disks. Setting the Returned bit
351 * ensures that this gets done only once -- we don't ever want to
352 * return -EIO here, instead we'll wait */
353
354 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
355 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
356 /* Maybe we can return now */
357 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
358 struct bio *mbio = r1_bio->master_bio;
359 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
360 (unsigned long long) mbio->bi_sector,
361 (unsigned long long) mbio->bi_sector +
362 (mbio->bi_size >> 9) - 1);
363 bio_endio(mbio, 0);
364 }
365 }
366 }
367 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
368 }
369 /*
370 *
371 * Let's see if all mirrored write operations have finished
372 * already.
373 */
374 if (atomic_dec_and_test(&r1_bio->remaining)) {
375 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state))
376 reschedule_retry(r1_bio);
377 else {
378 /* it really is the end of this request */
379 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
380 /* free extra copy of the data pages */
381 int i = bio->bi_vcnt;
382 while (i--)
383 safe_put_page(bio->bi_io_vec[i].bv_page);
384 }
385 /* clear the bitmap if all writes complete successfully */
386 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
387 r1_bio->sectors,
388 !test_bit(R1BIO_Degraded, &r1_bio->state),
389 behind);
390 md_write_end(r1_bio->mddev);
391 raid_end_bio_io(r1_bio);
392 }
393 }
394
395 if (to_put)
396 bio_put(to_put);
397 }
398
399
400 /*
401 * This routine returns the disk from which the requested read should
402 * be done. There is a per-array 'next expected sequential IO' sector
403 * number - if this matches on the next IO then we use the last disk.
404 * There is also a per-disk 'last know head position' sector that is
405 * maintained from IRQ contexts, both the normal and the resync IO
406 * completion handlers update this position correctly. If there is no
407 * perfect sequential match then we pick the disk whose head is closest.
408 *
409 * If there are 2 mirrors in the same 2 devices, performance degrades
410 * because position is mirror, not device based.
411 *
412 * The rdev for the device selected will have nr_pending incremented.
413 */
414 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
415 {
416 const unsigned long this_sector = r1_bio->sector;
417 int new_disk = conf->last_used, disk = new_disk;
418 int wonly_disk = -1;
419 const int sectors = r1_bio->sectors;
420 sector_t new_distance, current_distance;
421 mdk_rdev_t *rdev;
422
423 rcu_read_lock();
424 /*
425 * Check if we can balance. We can balance on the whole
426 * device if no resync is going on, or below the resync window.
427 * We take the first readable disk when above the resync window.
428 */
429 retry:
430 if (conf->mddev->recovery_cp < MaxSector &&
431 (this_sector + sectors >= conf->next_resync)) {
432 /* Choose the first operation device, for consistancy */
433 new_disk = 0;
434
435 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
436 r1_bio->bios[new_disk] == IO_BLOCKED ||
437 !rdev || !test_bit(In_sync, &rdev->flags)
438 || test_bit(WriteMostly, &rdev->flags);
439 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
440
441 if (rdev && test_bit(In_sync, &rdev->flags) &&
442 r1_bio->bios[new_disk] != IO_BLOCKED)
443 wonly_disk = new_disk;
444
445 if (new_disk == conf->raid_disks - 1) {
446 new_disk = wonly_disk;
447 break;
448 }
449 }
450 goto rb_out;
451 }
452
453
454 /* make sure the disk is operational */
455 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
456 r1_bio->bios[new_disk] == IO_BLOCKED ||
457 !rdev || !test_bit(In_sync, &rdev->flags) ||
458 test_bit(WriteMostly, &rdev->flags);
459 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
460
461 if (rdev && test_bit(In_sync, &rdev->flags) &&
462 r1_bio->bios[new_disk] != IO_BLOCKED)
463 wonly_disk = new_disk;
464
465 if (new_disk <= 0)
466 new_disk = conf->raid_disks;
467 new_disk--;
468 if (new_disk == disk) {
469 new_disk = wonly_disk;
470 break;
471 }
472 }
473
474 if (new_disk < 0)
475 goto rb_out;
476
477 disk = new_disk;
478 /* now disk == new_disk == starting point for search */
479
480 /*
481 * Don't change to another disk for sequential reads:
482 */
483 if (conf->next_seq_sect == this_sector)
484 goto rb_out;
485 if (this_sector == conf->mirrors[new_disk].head_position)
486 goto rb_out;
487
488 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
489
490 /* Find the disk whose head is closest */
491
492 do {
493 if (disk <= 0)
494 disk = conf->raid_disks;
495 disk--;
496
497 rdev = rcu_dereference(conf->mirrors[disk].rdev);
498
499 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
500 !test_bit(In_sync, &rdev->flags) ||
501 test_bit(WriteMostly, &rdev->flags))
502 continue;
503
504 if (!atomic_read(&rdev->nr_pending)) {
505 new_disk = disk;
506 break;
507 }
508 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
509 if (new_distance < current_distance) {
510 current_distance = new_distance;
511 new_disk = disk;
512 }
513 } while (disk != conf->last_used);
514
515 rb_out:
516
517
518 if (new_disk >= 0) {
519 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
520 if (!rdev)
521 goto retry;
522 atomic_inc(&rdev->nr_pending);
523 if (!test_bit(In_sync, &rdev->flags)) {
524 /* cannot risk returning a device that failed
525 * before we inc'ed nr_pending
526 */
527 rdev_dec_pending(rdev, conf->mddev);
528 goto retry;
529 }
530 conf->next_seq_sect = this_sector + sectors;
531 conf->last_used = new_disk;
532 }
533 rcu_read_unlock();
534
535 return new_disk;
536 }
537
538 static void unplug_slaves(mddev_t *mddev)
539 {
540 conf_t *conf = mddev_to_conf(mddev);
541 int i;
542
543 rcu_read_lock();
544 for (i=0; i<mddev->raid_disks; i++) {
545 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
546 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
547 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
548
549 atomic_inc(&rdev->nr_pending);
550 rcu_read_unlock();
551
552 if (r_queue->unplug_fn)
553 r_queue->unplug_fn(r_queue);
554
555 rdev_dec_pending(rdev, mddev);
556 rcu_read_lock();
557 }
558 }
559 rcu_read_unlock();
560 }
561
562 static void raid1_unplug(struct request_queue *q)
563 {
564 mddev_t *mddev = q->queuedata;
565
566 unplug_slaves(mddev);
567 md_wakeup_thread(mddev->thread);
568 }
569
570 static int raid1_issue_flush(struct request_queue *q, struct gendisk *disk,
571 sector_t *error_sector)
572 {
573 mddev_t *mddev = q->queuedata;
574 conf_t *conf = mddev_to_conf(mddev);
575 int i, ret = 0;
576
577 rcu_read_lock();
578 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
579 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
580 if (rdev && !test_bit(Faulty, &rdev->flags)) {
581 struct block_device *bdev = rdev->bdev;
582 struct request_queue *r_queue = bdev_get_queue(bdev);
583
584 if (!r_queue->issue_flush_fn)
585 ret = -EOPNOTSUPP;
586 else {
587 atomic_inc(&rdev->nr_pending);
588 rcu_read_unlock();
589 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
590 error_sector);
591 rdev_dec_pending(rdev, mddev);
592 rcu_read_lock();
593 }
594 }
595 }
596 rcu_read_unlock();
597 return ret;
598 }
599
600 static int raid1_congested(void *data, int bits)
601 {
602 mddev_t *mddev = data;
603 conf_t *conf = mddev_to_conf(mddev);
604 int i, ret = 0;
605
606 rcu_read_lock();
607 for (i = 0; i < mddev->raid_disks; i++) {
608 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
609 if (rdev && !test_bit(Faulty, &rdev->flags)) {
610 struct request_queue *q = bdev_get_queue(rdev->bdev);
611
612 /* Note the '|| 1' - when read_balance prefers
613 * non-congested targets, it can be removed
614 */
615 if ((bits & (1<<BDI_write_congested)) || 1)
616 ret |= bdi_congested(&q->backing_dev_info, bits);
617 else
618 ret &= bdi_congested(&q->backing_dev_info, bits);
619 }
620 }
621 rcu_read_unlock();
622 return ret;
623 }
624
625
626 /* Barriers....
627 * Sometimes we need to suspend IO while we do something else,
628 * either some resync/recovery, or reconfigure the array.
629 * To do this we raise a 'barrier'.
630 * The 'barrier' is a counter that can be raised multiple times
631 * to count how many activities are happening which preclude
632 * normal IO.
633 * We can only raise the barrier if there is no pending IO.
634 * i.e. if nr_pending == 0.
635 * We choose only to raise the barrier if no-one is waiting for the
636 * barrier to go down. This means that as soon as an IO request
637 * is ready, no other operations which require a barrier will start
638 * until the IO request has had a chance.
639 *
640 * So: regular IO calls 'wait_barrier'. When that returns there
641 * is no backgroup IO happening, It must arrange to call
642 * allow_barrier when it has finished its IO.
643 * backgroup IO calls must call raise_barrier. Once that returns
644 * there is no normal IO happeing. It must arrange to call
645 * lower_barrier when the particular background IO completes.
646 */
647 #define RESYNC_DEPTH 32
648
649 static void raise_barrier(conf_t *conf)
650 {
651 spin_lock_irq(&conf->resync_lock);
652
653 /* Wait until no block IO is waiting */
654 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
655 conf->resync_lock,
656 raid1_unplug(conf->mddev->queue));
657
658 /* block any new IO from starting */
659 conf->barrier++;
660
661 /* No wait for all pending IO to complete */
662 wait_event_lock_irq(conf->wait_barrier,
663 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
664 conf->resync_lock,
665 raid1_unplug(conf->mddev->queue));
666
667 spin_unlock_irq(&conf->resync_lock);
668 }
669
670 static void lower_barrier(conf_t *conf)
671 {
672 unsigned long flags;
673 spin_lock_irqsave(&conf->resync_lock, flags);
674 conf->barrier--;
675 spin_unlock_irqrestore(&conf->resync_lock, flags);
676 wake_up(&conf->wait_barrier);
677 }
678
679 static void wait_barrier(conf_t *conf)
680 {
681 spin_lock_irq(&conf->resync_lock);
682 if (conf->barrier) {
683 conf->nr_waiting++;
684 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
685 conf->resync_lock,
686 raid1_unplug(conf->mddev->queue));
687 conf->nr_waiting--;
688 }
689 conf->nr_pending++;
690 spin_unlock_irq(&conf->resync_lock);
691 }
692
693 static void allow_barrier(conf_t *conf)
694 {
695 unsigned long flags;
696 spin_lock_irqsave(&conf->resync_lock, flags);
697 conf->nr_pending--;
698 spin_unlock_irqrestore(&conf->resync_lock, flags);
699 wake_up(&conf->wait_barrier);
700 }
701
702 static void freeze_array(conf_t *conf)
703 {
704 /* stop syncio and normal IO and wait for everything to
705 * go quite.
706 * We increment barrier and nr_waiting, and then
707 * wait until barrier+nr_pending match nr_queued+2
708 */
709 spin_lock_irq(&conf->resync_lock);
710 conf->barrier++;
711 conf->nr_waiting++;
712 wait_event_lock_irq(conf->wait_barrier,
713 conf->barrier+conf->nr_pending == conf->nr_queued+2,
714 conf->resync_lock,
715 raid1_unplug(conf->mddev->queue));
716 spin_unlock_irq(&conf->resync_lock);
717 }
718 static void unfreeze_array(conf_t *conf)
719 {
720 /* reverse the effect of the freeze */
721 spin_lock_irq(&conf->resync_lock);
722 conf->barrier--;
723 conf->nr_waiting--;
724 wake_up(&conf->wait_barrier);
725 spin_unlock_irq(&conf->resync_lock);
726 }
727
728
729 /* duplicate the data pages for behind I/O */
730 static struct page **alloc_behind_pages(struct bio *bio)
731 {
732 int i;
733 struct bio_vec *bvec;
734 struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *),
735 GFP_NOIO);
736 if (unlikely(!pages))
737 goto do_sync_io;
738
739 bio_for_each_segment(bvec, bio, i) {
740 pages[i] = alloc_page(GFP_NOIO);
741 if (unlikely(!pages[i]))
742 goto do_sync_io;
743 memcpy(kmap(pages[i]) + bvec->bv_offset,
744 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
745 kunmap(pages[i]);
746 kunmap(bvec->bv_page);
747 }
748
749 return pages;
750
751 do_sync_io:
752 if (pages)
753 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
754 put_page(pages[i]);
755 kfree(pages);
756 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
757 return NULL;
758 }
759
760 static int make_request(struct request_queue *q, struct bio * bio)
761 {
762 mddev_t *mddev = q->queuedata;
763 conf_t *conf = mddev_to_conf(mddev);
764 mirror_info_t *mirror;
765 r1bio_t *r1_bio;
766 struct bio *read_bio;
767 int i, targets = 0, disks;
768 mdk_rdev_t *rdev;
769 struct bitmap *bitmap = mddev->bitmap;
770 unsigned long flags;
771 struct bio_list bl;
772 struct page **behind_pages = NULL;
773 const int rw = bio_data_dir(bio);
774 const int do_sync = bio_sync(bio);
775 int do_barriers;
776
777 /*
778 * Register the new request and wait if the reconstruction
779 * thread has put up a bar for new requests.
780 * Continue immediately if no resync is active currently.
781 * We test barriers_work *after* md_write_start as md_write_start
782 * may cause the first superblock write, and that will check out
783 * if barriers work.
784 */
785
786 md_write_start(mddev, bio); /* wait on superblock update early */
787
788 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
789 if (rw == WRITE)
790 md_write_end(mddev);
791 bio_endio(bio, -EOPNOTSUPP);
792 return 0;
793 }
794
795 wait_barrier(conf);
796
797 disk_stat_inc(mddev->gendisk, ios[rw]);
798 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
799
800 /*
801 * make_request() can abort the operation when READA is being
802 * used and no empty request is available.
803 *
804 */
805 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
806
807 r1_bio->master_bio = bio;
808 r1_bio->sectors = bio->bi_size >> 9;
809 r1_bio->state = 0;
810 r1_bio->mddev = mddev;
811 r1_bio->sector = bio->bi_sector;
812
813 if (rw == READ) {
814 /*
815 * read balancing logic:
816 */
817 int rdisk = read_balance(conf, r1_bio);
818
819 if (rdisk < 0) {
820 /* couldn't find anywhere to read from */
821 raid_end_bio_io(r1_bio);
822 return 0;
823 }
824 mirror = conf->mirrors + rdisk;
825
826 r1_bio->read_disk = rdisk;
827
828 read_bio = bio_clone(bio, GFP_NOIO);
829
830 r1_bio->bios[rdisk] = read_bio;
831
832 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
833 read_bio->bi_bdev = mirror->rdev->bdev;
834 read_bio->bi_end_io = raid1_end_read_request;
835 read_bio->bi_rw = READ | do_sync;
836 read_bio->bi_private = r1_bio;
837
838 generic_make_request(read_bio);
839 return 0;
840 }
841
842 /*
843 * WRITE:
844 */
845 /* first select target devices under spinlock and
846 * inc refcount on their rdev. Record them by setting
847 * bios[x] to bio
848 */
849 disks = conf->raid_disks;
850 #if 0
851 { static int first=1;
852 if (first) printk("First Write sector %llu disks %d\n",
853 (unsigned long long)r1_bio->sector, disks);
854 first = 0;
855 }
856 #endif
857 rcu_read_lock();
858 for (i = 0; i < disks; i++) {
859 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
860 !test_bit(Faulty, &rdev->flags)) {
861 atomic_inc(&rdev->nr_pending);
862 if (test_bit(Faulty, &rdev->flags)) {
863 rdev_dec_pending(rdev, mddev);
864 r1_bio->bios[i] = NULL;
865 } else
866 r1_bio->bios[i] = bio;
867 targets++;
868 } else
869 r1_bio->bios[i] = NULL;
870 }
871 rcu_read_unlock();
872
873 BUG_ON(targets == 0); /* we never fail the last device */
874
875 if (targets < conf->raid_disks) {
876 /* array is degraded, we will not clear the bitmap
877 * on I/O completion (see raid1_end_write_request) */
878 set_bit(R1BIO_Degraded, &r1_bio->state);
879 }
880
881 /* do behind I/O ? */
882 if (bitmap &&
883 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
884 (behind_pages = alloc_behind_pages(bio)) != NULL)
885 set_bit(R1BIO_BehindIO, &r1_bio->state);
886
887 atomic_set(&r1_bio->remaining, 0);
888 atomic_set(&r1_bio->behind_remaining, 0);
889
890 do_barriers = bio_barrier(bio);
891 if (do_barriers)
892 set_bit(R1BIO_Barrier, &r1_bio->state);
893
894 bio_list_init(&bl);
895 for (i = 0; i < disks; i++) {
896 struct bio *mbio;
897 if (!r1_bio->bios[i])
898 continue;
899
900 mbio = bio_clone(bio, GFP_NOIO);
901 r1_bio->bios[i] = mbio;
902
903 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
904 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
905 mbio->bi_end_io = raid1_end_write_request;
906 mbio->bi_rw = WRITE | do_barriers | do_sync;
907 mbio->bi_private = r1_bio;
908
909 if (behind_pages) {
910 struct bio_vec *bvec;
911 int j;
912
913 /* Yes, I really want the '__' version so that
914 * we clear any unused pointer in the io_vec, rather
915 * than leave them unchanged. This is important
916 * because when we come to free the pages, we won't
917 * know the originial bi_idx, so we just free
918 * them all
919 */
920 __bio_for_each_segment(bvec, mbio, j, 0)
921 bvec->bv_page = behind_pages[j];
922 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
923 atomic_inc(&r1_bio->behind_remaining);
924 }
925
926 atomic_inc(&r1_bio->remaining);
927
928 bio_list_add(&bl, mbio);
929 }
930 kfree(behind_pages); /* the behind pages are attached to the bios now */
931
932 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
933 test_bit(R1BIO_BehindIO, &r1_bio->state));
934 spin_lock_irqsave(&conf->device_lock, flags);
935 bio_list_merge(&conf->pending_bio_list, &bl);
936 bio_list_init(&bl);
937
938 blk_plug_device(mddev->queue);
939 spin_unlock_irqrestore(&conf->device_lock, flags);
940
941 if (do_sync)
942 md_wakeup_thread(mddev->thread);
943 #if 0
944 while ((bio = bio_list_pop(&bl)) != NULL)
945 generic_make_request(bio);
946 #endif
947
948 return 0;
949 }
950
951 static void status(struct seq_file *seq, mddev_t *mddev)
952 {
953 conf_t *conf = mddev_to_conf(mddev);
954 int i;
955
956 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
957 conf->raid_disks - mddev->degraded);
958 rcu_read_lock();
959 for (i = 0; i < conf->raid_disks; i++) {
960 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
961 seq_printf(seq, "%s",
962 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
963 }
964 rcu_read_unlock();
965 seq_printf(seq, "]");
966 }
967
968
969 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
970 {
971 char b[BDEVNAME_SIZE];
972 conf_t *conf = mddev_to_conf(mddev);
973
974 /*
975 * If it is not operational, then we have already marked it as dead
976 * else if it is the last working disks, ignore the error, let the
977 * next level up know.
978 * else mark the drive as failed
979 */
980 if (test_bit(In_sync, &rdev->flags)
981 && (conf->raid_disks - mddev->degraded) == 1)
982 /*
983 * Don't fail the drive, act as though we were just a
984 * normal single drive
985 */
986 return;
987 if (test_and_clear_bit(In_sync, &rdev->flags)) {
988 unsigned long flags;
989 spin_lock_irqsave(&conf->device_lock, flags);
990 mddev->degraded++;
991 set_bit(Faulty, &rdev->flags);
992 spin_unlock_irqrestore(&conf->device_lock, flags);
993 /*
994 * if recovery is running, make sure it aborts.
995 */
996 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
997 } else
998 set_bit(Faulty, &rdev->flags);
999 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1000 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
1001 " Operation continuing on %d devices\n",
1002 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
1003 }
1004
1005 static void print_conf(conf_t *conf)
1006 {
1007 int i;
1008
1009 printk("RAID1 conf printout:\n");
1010 if (!conf) {
1011 printk("(!conf)\n");
1012 return;
1013 }
1014 printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1015 conf->raid_disks);
1016
1017 rcu_read_lock();
1018 for (i = 0; i < conf->raid_disks; i++) {
1019 char b[BDEVNAME_SIZE];
1020 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
1021 if (rdev)
1022 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
1023 i, !test_bit(In_sync, &rdev->flags),
1024 !test_bit(Faulty, &rdev->flags),
1025 bdevname(rdev->bdev,b));
1026 }
1027 rcu_read_unlock();
1028 }
1029
1030 static void close_sync(conf_t *conf)
1031 {
1032 wait_barrier(conf);
1033 allow_barrier(conf);
1034
1035 mempool_destroy(conf->r1buf_pool);
1036 conf->r1buf_pool = NULL;
1037 }
1038
1039 static int raid1_spare_active(mddev_t *mddev)
1040 {
1041 int i;
1042 conf_t *conf = mddev->private;
1043
1044 /*
1045 * Find all failed disks within the RAID1 configuration
1046 * and mark them readable.
1047 * Called under mddev lock, so rcu protection not needed.
1048 */
1049 for (i = 0; i < conf->raid_disks; i++) {
1050 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
1051 if (rdev
1052 && !test_bit(Faulty, &rdev->flags)
1053 && !test_and_set_bit(In_sync, &rdev->flags)) {
1054 unsigned long flags;
1055 spin_lock_irqsave(&conf->device_lock, flags);
1056 mddev->degraded--;
1057 spin_unlock_irqrestore(&conf->device_lock, flags);
1058 }
1059 }
1060
1061 print_conf(conf);
1062 return 0;
1063 }
1064
1065
1066 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1067 {
1068 conf_t *conf = mddev->private;
1069 int found = 0;
1070 int mirror = 0;
1071 mirror_info_t *p;
1072
1073 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1074 if ( !(p=conf->mirrors+mirror)->rdev) {
1075
1076 blk_queue_stack_limits(mddev->queue,
1077 rdev->bdev->bd_disk->queue);
1078 /* as we don't honour merge_bvec_fn, we must never risk
1079 * violating it, so limit ->max_sector to one PAGE, as
1080 * a one page request is never in violation.
1081 */
1082 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1083 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1084 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1085
1086 p->head_position = 0;
1087 rdev->raid_disk = mirror;
1088 found = 1;
1089 /* As all devices are equivalent, we don't need a full recovery
1090 * if this was recently any drive of the array
1091 */
1092 if (rdev->saved_raid_disk < 0)
1093 conf->fullsync = 1;
1094 rcu_assign_pointer(p->rdev, rdev);
1095 break;
1096 }
1097
1098 print_conf(conf);
1099 return found;
1100 }
1101
1102 static int raid1_remove_disk(mddev_t *mddev, int number)
1103 {
1104 conf_t *conf = mddev->private;
1105 int err = 0;
1106 mdk_rdev_t *rdev;
1107 mirror_info_t *p = conf->mirrors+ number;
1108
1109 print_conf(conf);
1110 rdev = p->rdev;
1111 if (rdev) {
1112 if (test_bit(In_sync, &rdev->flags) ||
1113 atomic_read(&rdev->nr_pending)) {
1114 err = -EBUSY;
1115 goto abort;
1116 }
1117 p->rdev = NULL;
1118 synchronize_rcu();
1119 if (atomic_read(&rdev->nr_pending)) {
1120 /* lost the race, try later */
1121 err = -EBUSY;
1122 p->rdev = rdev;
1123 }
1124 }
1125 abort:
1126
1127 print_conf(conf);
1128 return err;
1129 }
1130
1131
1132 static void end_sync_read(struct bio *bio, int error)
1133 {
1134 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1135 int i;
1136
1137 for (i=r1_bio->mddev->raid_disks; i--; )
1138 if (r1_bio->bios[i] == bio)
1139 break;
1140 BUG_ON(i < 0);
1141 update_head_pos(i, r1_bio);
1142 /*
1143 * we have read a block, now it needs to be re-written,
1144 * or re-read if the read failed.
1145 * We don't do much here, just schedule handling by raid1d
1146 */
1147 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1148 set_bit(R1BIO_Uptodate, &r1_bio->state);
1149
1150 if (atomic_dec_and_test(&r1_bio->remaining))
1151 reschedule_retry(r1_bio);
1152 }
1153
1154 static void end_sync_write(struct bio *bio, int error)
1155 {
1156 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1157 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1158 mddev_t *mddev = r1_bio->mddev;
1159 conf_t *conf = mddev_to_conf(mddev);
1160 int i;
1161 int mirror=0;
1162
1163 for (i = 0; i < conf->raid_disks; i++)
1164 if (r1_bio->bios[i] == bio) {
1165 mirror = i;
1166 break;
1167 }
1168 if (!uptodate) {
1169 int sync_blocks = 0;
1170 sector_t s = r1_bio->sector;
1171 long sectors_to_go = r1_bio->sectors;
1172 /* make sure these bits doesn't get cleared. */
1173 do {
1174 bitmap_end_sync(mddev->bitmap, s,
1175 &sync_blocks, 1);
1176 s += sync_blocks;
1177 sectors_to_go -= sync_blocks;
1178 } while (sectors_to_go > 0);
1179 md_error(mddev, conf->mirrors[mirror].rdev);
1180 }
1181
1182 update_head_pos(mirror, r1_bio);
1183
1184 if (atomic_dec_and_test(&r1_bio->remaining)) {
1185 md_done_sync(mddev, r1_bio->sectors, uptodate);
1186 put_buf(r1_bio);
1187 }
1188 }
1189
1190 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1191 {
1192 conf_t *conf = mddev_to_conf(mddev);
1193 int i;
1194 int disks = conf->raid_disks;
1195 struct bio *bio, *wbio;
1196
1197 bio = r1_bio->bios[r1_bio->read_disk];
1198
1199
1200 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1201 /* We have read all readable devices. If we haven't
1202 * got the block, then there is no hope left.
1203 * If we have, then we want to do a comparison
1204 * and skip the write if everything is the same.
1205 * If any blocks failed to read, then we need to
1206 * attempt an over-write
1207 */
1208 int primary;
1209 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1210 for (i=0; i<mddev->raid_disks; i++)
1211 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1212 md_error(mddev, conf->mirrors[i].rdev);
1213
1214 md_done_sync(mddev, r1_bio->sectors, 1);
1215 put_buf(r1_bio);
1216 return;
1217 }
1218 for (primary=0; primary<mddev->raid_disks; primary++)
1219 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1220 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1221 r1_bio->bios[primary]->bi_end_io = NULL;
1222 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1223 break;
1224 }
1225 r1_bio->read_disk = primary;
1226 for (i=0; i<mddev->raid_disks; i++)
1227 if (r1_bio->bios[i]->bi_end_io == end_sync_read) {
1228 int j;
1229 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1230 struct bio *pbio = r1_bio->bios[primary];
1231 struct bio *sbio = r1_bio->bios[i];
1232
1233 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1234 for (j = vcnt; j-- ; ) {
1235 struct page *p, *s;
1236 p = pbio->bi_io_vec[j].bv_page;
1237 s = sbio->bi_io_vec[j].bv_page;
1238 if (memcmp(page_address(p),
1239 page_address(s),
1240 PAGE_SIZE))
1241 break;
1242 }
1243 } else
1244 j = 0;
1245 if (j >= 0)
1246 mddev->resync_mismatches += r1_bio->sectors;
1247 if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
1248 sbio->bi_end_io = NULL;
1249 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1250 } else {
1251 /* fixup the bio for reuse */
1252 sbio->bi_vcnt = vcnt;
1253 sbio->bi_size = r1_bio->sectors << 9;
1254 sbio->bi_idx = 0;
1255 sbio->bi_phys_segments = 0;
1256 sbio->bi_hw_segments = 0;
1257 sbio->bi_hw_front_size = 0;
1258 sbio->bi_hw_back_size = 0;
1259 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1260 sbio->bi_flags |= 1 << BIO_UPTODATE;
1261 sbio->bi_next = NULL;
1262 sbio->bi_sector = r1_bio->sector +
1263 conf->mirrors[i].rdev->data_offset;
1264 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1265 for (j = 0; j < vcnt ; j++)
1266 memcpy(page_address(sbio->bi_io_vec[j].bv_page),
1267 page_address(pbio->bi_io_vec[j].bv_page),
1268 PAGE_SIZE);
1269
1270 }
1271 }
1272 }
1273 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1274 /* ouch - failed to read all of that.
1275 * Try some synchronous reads of other devices to get
1276 * good data, much like with normal read errors. Only
1277 * read into the pages we already have so we don't
1278 * need to re-issue the read request.
1279 * We don't need to freeze the array, because being in an
1280 * active sync request, there is no normal IO, and
1281 * no overlapping syncs.
1282 */
1283 sector_t sect = r1_bio->sector;
1284 int sectors = r1_bio->sectors;
1285 int idx = 0;
1286
1287 while(sectors) {
1288 int s = sectors;
1289 int d = r1_bio->read_disk;
1290 int success = 0;
1291 mdk_rdev_t *rdev;
1292
1293 if (s > (PAGE_SIZE>>9))
1294 s = PAGE_SIZE >> 9;
1295 do {
1296 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1297 /* No rcu protection needed here devices
1298 * can only be removed when no resync is
1299 * active, and resync is currently active
1300 */
1301 rdev = conf->mirrors[d].rdev;
1302 if (sync_page_io(rdev->bdev,
1303 sect + rdev->data_offset,
1304 s<<9,
1305 bio->bi_io_vec[idx].bv_page,
1306 READ)) {
1307 success = 1;
1308 break;
1309 }
1310 }
1311 d++;
1312 if (d == conf->raid_disks)
1313 d = 0;
1314 } while (!success && d != r1_bio->read_disk);
1315
1316 if (success) {
1317 int start = d;
1318 /* write it back and re-read */
1319 set_bit(R1BIO_Uptodate, &r1_bio->state);
1320 while (d != r1_bio->read_disk) {
1321 if (d == 0)
1322 d = conf->raid_disks;
1323 d--;
1324 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1325 continue;
1326 rdev = conf->mirrors[d].rdev;
1327 atomic_add(s, &rdev->corrected_errors);
1328 if (sync_page_io(rdev->bdev,
1329 sect + rdev->data_offset,
1330 s<<9,
1331 bio->bi_io_vec[idx].bv_page,
1332 WRITE) == 0)
1333 md_error(mddev, rdev);
1334 }
1335 d = start;
1336 while (d != r1_bio->read_disk) {
1337 if (d == 0)
1338 d = conf->raid_disks;
1339 d--;
1340 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1341 continue;
1342 rdev = conf->mirrors[d].rdev;
1343 if (sync_page_io(rdev->bdev,
1344 sect + rdev->data_offset,
1345 s<<9,
1346 bio->bi_io_vec[idx].bv_page,
1347 READ) == 0)
1348 md_error(mddev, rdev);
1349 }
1350 } else {
1351 char b[BDEVNAME_SIZE];
1352 /* Cannot read from anywhere, array is toast */
1353 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1354 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1355 " for block %llu\n",
1356 bdevname(bio->bi_bdev,b),
1357 (unsigned long long)r1_bio->sector);
1358 md_done_sync(mddev, r1_bio->sectors, 0);
1359 put_buf(r1_bio);
1360 return;
1361 }
1362 sectors -= s;
1363 sect += s;
1364 idx ++;
1365 }
1366 }
1367
1368 /*
1369 * schedule writes
1370 */
1371 atomic_set(&r1_bio->remaining, 1);
1372 for (i = 0; i < disks ; i++) {
1373 wbio = r1_bio->bios[i];
1374 if (wbio->bi_end_io == NULL ||
1375 (wbio->bi_end_io == end_sync_read &&
1376 (i == r1_bio->read_disk ||
1377 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1378 continue;
1379
1380 wbio->bi_rw = WRITE;
1381 wbio->bi_end_io = end_sync_write;
1382 atomic_inc(&r1_bio->remaining);
1383 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
1384
1385 generic_make_request(wbio);
1386 }
1387
1388 if (atomic_dec_and_test(&r1_bio->remaining)) {
1389 /* if we're here, all write(s) have completed, so clean up */
1390 md_done_sync(mddev, r1_bio->sectors, 1);
1391 put_buf(r1_bio);
1392 }
1393 }
1394
1395 /*
1396 * This is a kernel thread which:
1397 *
1398 * 1. Retries failed read operations on working mirrors.
1399 * 2. Updates the raid superblock when problems encounter.
1400 * 3. Performs writes following reads for array syncronising.
1401 */
1402
1403 static void fix_read_error(conf_t *conf, int read_disk,
1404 sector_t sect, int sectors)
1405 {
1406 mddev_t *mddev = conf->mddev;
1407 while(sectors) {
1408 int s = sectors;
1409 int d = read_disk;
1410 int success = 0;
1411 int start;
1412 mdk_rdev_t *rdev;
1413
1414 if (s > (PAGE_SIZE>>9))
1415 s = PAGE_SIZE >> 9;
1416
1417 do {
1418 /* Note: no rcu protection needed here
1419 * as this is synchronous in the raid1d thread
1420 * which is the thread that might remove
1421 * a device. If raid1d ever becomes multi-threaded....
1422 */
1423 rdev = conf->mirrors[d].rdev;
1424 if (rdev &&
1425 test_bit(In_sync, &rdev->flags) &&
1426 sync_page_io(rdev->bdev,
1427 sect + rdev->data_offset,
1428 s<<9,
1429 conf->tmppage, READ))
1430 success = 1;
1431 else {
1432 d++;
1433 if (d == conf->raid_disks)
1434 d = 0;
1435 }
1436 } while (!success && d != read_disk);
1437
1438 if (!success) {
1439 /* Cannot read from anywhere -- bye bye array */
1440 md_error(mddev, conf->mirrors[read_disk].rdev);
1441 break;
1442 }
1443 /* write it back and re-read */
1444 start = d;
1445 while (d != read_disk) {
1446 if (d==0)
1447 d = conf->raid_disks;
1448 d--;
1449 rdev = conf->mirrors[d].rdev;
1450 if (rdev &&
1451 test_bit(In_sync, &rdev->flags)) {
1452 if (sync_page_io(rdev->bdev,
1453 sect + rdev->data_offset,
1454 s<<9, conf->tmppage, WRITE)
1455 == 0)
1456 /* Well, this device is dead */
1457 md_error(mddev, rdev);
1458 }
1459 }
1460 d = start;
1461 while (d != read_disk) {
1462 char b[BDEVNAME_SIZE];
1463 if (d==0)
1464 d = conf->raid_disks;
1465 d--;
1466 rdev = conf->mirrors[d].rdev;
1467 if (rdev &&
1468 test_bit(In_sync, &rdev->flags)) {
1469 if (sync_page_io(rdev->bdev,
1470 sect + rdev->data_offset,
1471 s<<9, conf->tmppage, READ)
1472 == 0)
1473 /* Well, this device is dead */
1474 md_error(mddev, rdev);
1475 else {
1476 atomic_add(s, &rdev->corrected_errors);
1477 printk(KERN_INFO
1478 "raid1:%s: read error corrected "
1479 "(%d sectors at %llu on %s)\n",
1480 mdname(mddev), s,
1481 (unsigned long long)(sect +
1482 rdev->data_offset),
1483 bdevname(rdev->bdev, b));
1484 }
1485 }
1486 }
1487 sectors -= s;
1488 sect += s;
1489 }
1490 }
1491
1492 static void raid1d(mddev_t *mddev)
1493 {
1494 r1bio_t *r1_bio;
1495 struct bio *bio;
1496 unsigned long flags;
1497 conf_t *conf = mddev_to_conf(mddev);
1498 struct list_head *head = &conf->retry_list;
1499 int unplug=0;
1500 mdk_rdev_t *rdev;
1501
1502 md_check_recovery(mddev);
1503
1504 for (;;) {
1505 char b[BDEVNAME_SIZE];
1506 spin_lock_irqsave(&conf->device_lock, flags);
1507
1508 if (conf->pending_bio_list.head) {
1509 bio = bio_list_get(&conf->pending_bio_list);
1510 blk_remove_plug(mddev->queue);
1511 spin_unlock_irqrestore(&conf->device_lock, flags);
1512 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1513 bitmap_unplug(mddev->bitmap);
1514
1515 while (bio) { /* submit pending writes */
1516 struct bio *next = bio->bi_next;
1517 bio->bi_next = NULL;
1518 generic_make_request(bio);
1519 bio = next;
1520 }
1521 unplug = 1;
1522
1523 continue;
1524 }
1525
1526 if (list_empty(head))
1527 break;
1528 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1529 list_del(head->prev);
1530 conf->nr_queued--;
1531 spin_unlock_irqrestore(&conf->device_lock, flags);
1532
1533 mddev = r1_bio->mddev;
1534 conf = mddev_to_conf(mddev);
1535 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1536 sync_request_write(mddev, r1_bio);
1537 unplug = 1;
1538 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1539 /* some requests in the r1bio were BIO_RW_BARRIER
1540 * requests which failed with -EOPNOTSUPP. Hohumm..
1541 * Better resubmit without the barrier.
1542 * We know which devices to resubmit for, because
1543 * all others have had their bios[] entry cleared.
1544 * We already have a nr_pending reference on these rdevs.
1545 */
1546 int i;
1547 const int do_sync = bio_sync(r1_bio->master_bio);
1548 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1549 clear_bit(R1BIO_Barrier, &r1_bio->state);
1550 for (i=0; i < conf->raid_disks; i++)
1551 if (r1_bio->bios[i])
1552 atomic_inc(&r1_bio->remaining);
1553 for (i=0; i < conf->raid_disks; i++)
1554 if (r1_bio->bios[i]) {
1555 struct bio_vec *bvec;
1556 int j;
1557
1558 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1559 /* copy pages from the failed bio, as
1560 * this might be a write-behind device */
1561 __bio_for_each_segment(bvec, bio, j, 0)
1562 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1563 bio_put(r1_bio->bios[i]);
1564 bio->bi_sector = r1_bio->sector +
1565 conf->mirrors[i].rdev->data_offset;
1566 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1567 bio->bi_end_io = raid1_end_write_request;
1568 bio->bi_rw = WRITE | do_sync;
1569 bio->bi_private = r1_bio;
1570 r1_bio->bios[i] = bio;
1571 generic_make_request(bio);
1572 }
1573 } else {
1574 int disk;
1575
1576 /* we got a read error. Maybe the drive is bad. Maybe just
1577 * the block and we can fix it.
1578 * We freeze all other IO, and try reading the block from
1579 * other devices. When we find one, we re-write
1580 * and check it that fixes the read error.
1581 * This is all done synchronously while the array is
1582 * frozen
1583 */
1584 if (mddev->ro == 0) {
1585 freeze_array(conf);
1586 fix_read_error(conf, r1_bio->read_disk,
1587 r1_bio->sector,
1588 r1_bio->sectors);
1589 unfreeze_array(conf);
1590 }
1591
1592 bio = r1_bio->bios[r1_bio->read_disk];
1593 if ((disk=read_balance(conf, r1_bio)) == -1) {
1594 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1595 " read error for block %llu\n",
1596 bdevname(bio->bi_bdev,b),
1597 (unsigned long long)r1_bio->sector);
1598 raid_end_bio_io(r1_bio);
1599 } else {
1600 const int do_sync = bio_sync(r1_bio->master_bio);
1601 r1_bio->bios[r1_bio->read_disk] =
1602 mddev->ro ? IO_BLOCKED : NULL;
1603 r1_bio->read_disk = disk;
1604 bio_put(bio);
1605 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1606 r1_bio->bios[r1_bio->read_disk] = bio;
1607 rdev = conf->mirrors[disk].rdev;
1608 if (printk_ratelimit())
1609 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1610 " another mirror\n",
1611 bdevname(rdev->bdev,b),
1612 (unsigned long long)r1_bio->sector);
1613 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1614 bio->bi_bdev = rdev->bdev;
1615 bio->bi_end_io = raid1_end_read_request;
1616 bio->bi_rw = READ | do_sync;
1617 bio->bi_private = r1_bio;
1618 unplug = 1;
1619 generic_make_request(bio);
1620 }
1621 }
1622 }
1623 spin_unlock_irqrestore(&conf->device_lock, flags);
1624 if (unplug)
1625 unplug_slaves(mddev);
1626 }
1627
1628
1629 static int init_resync(conf_t *conf)
1630 {
1631 int buffs;
1632
1633 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1634 BUG_ON(conf->r1buf_pool);
1635 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1636 conf->poolinfo);
1637 if (!conf->r1buf_pool)
1638 return -ENOMEM;
1639 conf->next_resync = 0;
1640 return 0;
1641 }
1642
1643 /*
1644 * perform a "sync" on one "block"
1645 *
1646 * We need to make sure that no normal I/O request - particularly write
1647 * requests - conflict with active sync requests.
1648 *
1649 * This is achieved by tracking pending requests and a 'barrier' concept
1650 * that can be installed to exclude normal IO requests.
1651 */
1652
1653 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1654 {
1655 conf_t *conf = mddev_to_conf(mddev);
1656 r1bio_t *r1_bio;
1657 struct bio *bio;
1658 sector_t max_sector, nr_sectors;
1659 int disk = -1;
1660 int i;
1661 int wonly = -1;
1662 int write_targets = 0, read_targets = 0;
1663 int sync_blocks;
1664 int still_degraded = 0;
1665
1666 if (!conf->r1buf_pool)
1667 {
1668 /*
1669 printk("sync start - bitmap %p\n", mddev->bitmap);
1670 */
1671 if (init_resync(conf))
1672 return 0;
1673 }
1674
1675 max_sector = mddev->size << 1;
1676 if (sector_nr >= max_sector) {
1677 /* If we aborted, we need to abort the
1678 * sync on the 'current' bitmap chunk (there will
1679 * only be one in raid1 resync.
1680 * We can find the current addess in mddev->curr_resync
1681 */
1682 if (mddev->curr_resync < max_sector) /* aborted */
1683 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1684 &sync_blocks, 1);
1685 else /* completed sync */
1686 conf->fullsync = 0;
1687
1688 bitmap_close_sync(mddev->bitmap);
1689 close_sync(conf);
1690 return 0;
1691 }
1692
1693 if (mddev->bitmap == NULL &&
1694 mddev->recovery_cp == MaxSector &&
1695 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
1696 conf->fullsync == 0) {
1697 *skipped = 1;
1698 return max_sector - sector_nr;
1699 }
1700 /* before building a request, check if we can skip these blocks..
1701 * This call the bitmap_start_sync doesn't actually record anything
1702 */
1703 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1704 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1705 /* We can skip this block, and probably several more */
1706 *skipped = 1;
1707 return sync_blocks;
1708 }
1709 /*
1710 * If there is non-resync activity waiting for a turn,
1711 * and resync is going fast enough,
1712 * then let it though before starting on this new sync request.
1713 */
1714 if (!go_faster && conf->nr_waiting)
1715 msleep_interruptible(1000);
1716
1717 raise_barrier(conf);
1718
1719 conf->next_resync = sector_nr;
1720
1721 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1722 rcu_read_lock();
1723 /*
1724 * If we get a correctably read error during resync or recovery,
1725 * we might want to read from a different device. So we
1726 * flag all drives that could conceivably be read from for READ,
1727 * and any others (which will be non-In_sync devices) for WRITE.
1728 * If a read fails, we try reading from something else for which READ
1729 * is OK.
1730 */
1731
1732 r1_bio->mddev = mddev;
1733 r1_bio->sector = sector_nr;
1734 r1_bio->state = 0;
1735 set_bit(R1BIO_IsSync, &r1_bio->state);
1736
1737 for (i=0; i < conf->raid_disks; i++) {
1738 mdk_rdev_t *rdev;
1739 bio = r1_bio->bios[i];
1740
1741 /* take from bio_init */
1742 bio->bi_next = NULL;
1743 bio->bi_flags |= 1 << BIO_UPTODATE;
1744 bio->bi_rw = READ;
1745 bio->bi_vcnt = 0;
1746 bio->bi_idx = 0;
1747 bio->bi_phys_segments = 0;
1748 bio->bi_hw_segments = 0;
1749 bio->bi_size = 0;
1750 bio->bi_end_io = NULL;
1751 bio->bi_private = NULL;
1752
1753 rdev = rcu_dereference(conf->mirrors[i].rdev);
1754 if (rdev == NULL ||
1755 test_bit(Faulty, &rdev->flags)) {
1756 still_degraded = 1;
1757 continue;
1758 } else if (!test_bit(In_sync, &rdev->flags)) {
1759 bio->bi_rw = WRITE;
1760 bio->bi_end_io = end_sync_write;
1761 write_targets ++;
1762 } else {
1763 /* may need to read from here */
1764 bio->bi_rw = READ;
1765 bio->bi_end_io = end_sync_read;
1766 if (test_bit(WriteMostly, &rdev->flags)) {
1767 if (wonly < 0)
1768 wonly = i;
1769 } else {
1770 if (disk < 0)
1771 disk = i;
1772 }
1773 read_targets++;
1774 }
1775 atomic_inc(&rdev->nr_pending);
1776 bio->bi_sector = sector_nr + rdev->data_offset;
1777 bio->bi_bdev = rdev->bdev;
1778 bio->bi_private = r1_bio;
1779 }
1780 rcu_read_unlock();
1781 if (disk < 0)
1782 disk = wonly;
1783 r1_bio->read_disk = disk;
1784
1785 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1786 /* extra read targets are also write targets */
1787 write_targets += read_targets-1;
1788
1789 if (write_targets == 0 || read_targets == 0) {
1790 /* There is nowhere to write, so all non-sync
1791 * drives must be failed - so we are finished
1792 */
1793 sector_t rv = max_sector - sector_nr;
1794 *skipped = 1;
1795 put_buf(r1_bio);
1796 return rv;
1797 }
1798
1799 nr_sectors = 0;
1800 sync_blocks = 0;
1801 do {
1802 struct page *page;
1803 int len = PAGE_SIZE;
1804 if (sector_nr + (len>>9) > max_sector)
1805 len = (max_sector - sector_nr) << 9;
1806 if (len == 0)
1807 break;
1808 if (sync_blocks == 0) {
1809 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1810 &sync_blocks, still_degraded) &&
1811 !conf->fullsync &&
1812 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1813 break;
1814 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
1815 if (len > (sync_blocks<<9))
1816 len = sync_blocks<<9;
1817 }
1818
1819 for (i=0 ; i < conf->raid_disks; i++) {
1820 bio = r1_bio->bios[i];
1821 if (bio->bi_end_io) {
1822 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
1823 if (bio_add_page(bio, page, len, 0) == 0) {
1824 /* stop here */
1825 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
1826 while (i > 0) {
1827 i--;
1828 bio = r1_bio->bios[i];
1829 if (bio->bi_end_io==NULL)
1830 continue;
1831 /* remove last page from this bio */
1832 bio->bi_vcnt--;
1833 bio->bi_size -= len;
1834 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1835 }
1836 goto bio_full;
1837 }
1838 }
1839 }
1840 nr_sectors += len>>9;
1841 sector_nr += len>>9;
1842 sync_blocks -= (len>>9);
1843 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1844 bio_full:
1845 r1_bio->sectors = nr_sectors;
1846
1847 /* For a user-requested sync, we read all readable devices and do a
1848 * compare
1849 */
1850 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1851 atomic_set(&r1_bio->remaining, read_targets);
1852 for (i=0; i<conf->raid_disks; i++) {
1853 bio = r1_bio->bios[i];
1854 if (bio->bi_end_io == end_sync_read) {
1855 md_sync_acct(bio->bi_bdev, nr_sectors);
1856 generic_make_request(bio);
1857 }
1858 }
1859 } else {
1860 atomic_set(&r1_bio->remaining, 1);
1861 bio = r1_bio->bios[r1_bio->read_disk];
1862 md_sync_acct(bio->bi_bdev, nr_sectors);
1863 generic_make_request(bio);
1864
1865 }
1866 return nr_sectors;
1867 }
1868
1869 static int run(mddev_t *mddev)
1870 {
1871 conf_t *conf;
1872 int i, j, disk_idx;
1873 mirror_info_t *disk;
1874 mdk_rdev_t *rdev;
1875 struct list_head *tmp;
1876
1877 if (mddev->level != 1) {
1878 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1879 mdname(mddev), mddev->level);
1880 goto out;
1881 }
1882 if (mddev->reshape_position != MaxSector) {
1883 printk("raid1: %s: reshape_position set but not supported\n",
1884 mdname(mddev));
1885 goto out;
1886 }
1887 /*
1888 * copy the already verified devices into our private RAID1
1889 * bookkeeping area. [whatever we allocate in run(),
1890 * should be freed in stop()]
1891 */
1892 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
1893 mddev->private = conf;
1894 if (!conf)
1895 goto out_no_mem;
1896
1897 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1898 GFP_KERNEL);
1899 if (!conf->mirrors)
1900 goto out_no_mem;
1901
1902 conf->tmppage = alloc_page(GFP_KERNEL);
1903 if (!conf->tmppage)
1904 goto out_no_mem;
1905
1906 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1907 if (!conf->poolinfo)
1908 goto out_no_mem;
1909 conf->poolinfo->mddev = mddev;
1910 conf->poolinfo->raid_disks = mddev->raid_disks;
1911 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1912 r1bio_pool_free,
1913 conf->poolinfo);
1914 if (!conf->r1bio_pool)
1915 goto out_no_mem;
1916
1917 ITERATE_RDEV(mddev, rdev, tmp) {
1918 disk_idx = rdev->raid_disk;
1919 if (disk_idx >= mddev->raid_disks
1920 || disk_idx < 0)
1921 continue;
1922 disk = conf->mirrors + disk_idx;
1923
1924 disk->rdev = rdev;
1925
1926 blk_queue_stack_limits(mddev->queue,
1927 rdev->bdev->bd_disk->queue);
1928 /* as we don't honour merge_bvec_fn, we must never risk
1929 * violating it, so limit ->max_sector to one PAGE, as
1930 * a one page request is never in violation.
1931 */
1932 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1933 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1934 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1935
1936 disk->head_position = 0;
1937 }
1938 conf->raid_disks = mddev->raid_disks;
1939 conf->mddev = mddev;
1940 spin_lock_init(&conf->device_lock);
1941 INIT_LIST_HEAD(&conf->retry_list);
1942
1943 spin_lock_init(&conf->resync_lock);
1944 init_waitqueue_head(&conf->wait_barrier);
1945
1946 bio_list_init(&conf->pending_bio_list);
1947 bio_list_init(&conf->flushing_bio_list);
1948
1949
1950 mddev->degraded = 0;
1951 for (i = 0; i < conf->raid_disks; i++) {
1952
1953 disk = conf->mirrors + i;
1954
1955 if (!disk->rdev ||
1956 !test_bit(In_sync, &disk->rdev->flags)) {
1957 disk->head_position = 0;
1958 mddev->degraded++;
1959 if (disk->rdev)
1960 conf->fullsync = 1;
1961 }
1962 }
1963 if (mddev->degraded == conf->raid_disks) {
1964 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1965 mdname(mddev));
1966 goto out_free_conf;
1967 }
1968 if (conf->raid_disks - mddev->degraded == 1)
1969 mddev->recovery_cp = MaxSector;
1970
1971 /*
1972 * find the first working one and use it as a starting point
1973 * to read balancing.
1974 */
1975 for (j = 0; j < conf->raid_disks &&
1976 (!conf->mirrors[j].rdev ||
1977 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
1978 /* nothing */;
1979 conf->last_used = j;
1980
1981
1982 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1983 if (!mddev->thread) {
1984 printk(KERN_ERR
1985 "raid1: couldn't allocate thread for %s\n",
1986 mdname(mddev));
1987 goto out_free_conf;
1988 }
1989
1990 printk(KERN_INFO
1991 "raid1: raid set %s active with %d out of %d mirrors\n",
1992 mdname(mddev), mddev->raid_disks - mddev->degraded,
1993 mddev->raid_disks);
1994 /*
1995 * Ok, everything is just fine now
1996 */
1997 mddev->array_size = mddev->size;
1998
1999 mddev->queue->unplug_fn = raid1_unplug;
2000 mddev->queue->issue_flush_fn = raid1_issue_flush;
2001 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2002 mddev->queue->backing_dev_info.congested_data = mddev;
2003
2004 return 0;
2005
2006 out_no_mem:
2007 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
2008 mdname(mddev));
2009
2010 out_free_conf:
2011 if (conf) {
2012 if (conf->r1bio_pool)
2013 mempool_destroy(conf->r1bio_pool);
2014 kfree(conf->mirrors);
2015 safe_put_page(conf->tmppage);
2016 kfree(conf->poolinfo);
2017 kfree(conf);
2018 mddev->private = NULL;
2019 }
2020 out:
2021 return -EIO;
2022 }
2023
2024 static int stop(mddev_t *mddev)
2025 {
2026 conf_t *conf = mddev_to_conf(mddev);
2027 struct bitmap *bitmap = mddev->bitmap;
2028 int behind_wait = 0;
2029
2030 /* wait for behind writes to complete */
2031 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
2032 behind_wait++;
2033 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
2034 set_current_state(TASK_UNINTERRUPTIBLE);
2035 schedule_timeout(HZ); /* wait a second */
2036 /* need to kick something here to make sure I/O goes? */
2037 }
2038
2039 md_unregister_thread(mddev->thread);
2040 mddev->thread = NULL;
2041 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
2042 if (conf->r1bio_pool)
2043 mempool_destroy(conf->r1bio_pool);
2044 kfree(conf->mirrors);
2045 kfree(conf->poolinfo);
2046 kfree(conf);
2047 mddev->private = NULL;
2048 return 0;
2049 }
2050
2051 static int raid1_resize(mddev_t *mddev, sector_t sectors)
2052 {
2053 /* no resync is happening, and there is enough space
2054 * on all devices, so we can resize.
2055 * We need to make sure resync covers any new space.
2056 * If the array is shrinking we should possibly wait until
2057 * any io in the removed space completes, but it hardly seems
2058 * worth it.
2059 */
2060 mddev->array_size = sectors>>1;
2061 set_capacity(mddev->gendisk, mddev->array_size << 1);
2062 mddev->changed = 1;
2063 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
2064 mddev->recovery_cp = mddev->size << 1;
2065 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2066 }
2067 mddev->size = mddev->array_size;
2068 mddev->resync_max_sectors = sectors;
2069 return 0;
2070 }
2071
2072 static int raid1_reshape(mddev_t *mddev)
2073 {
2074 /* We need to:
2075 * 1/ resize the r1bio_pool
2076 * 2/ resize conf->mirrors
2077 *
2078 * We allocate a new r1bio_pool if we can.
2079 * Then raise a device barrier and wait until all IO stops.
2080 * Then resize conf->mirrors and swap in the new r1bio pool.
2081 *
2082 * At the same time, we "pack" the devices so that all the missing
2083 * devices have the higher raid_disk numbers.
2084 */
2085 mempool_t *newpool, *oldpool;
2086 struct pool_info *newpoolinfo;
2087 mirror_info_t *newmirrors;
2088 conf_t *conf = mddev_to_conf(mddev);
2089 int cnt, raid_disks;
2090 unsigned long flags;
2091 int d, d2;
2092
2093 /* Cannot change chunk_size, layout, or level */
2094 if (mddev->chunk_size != mddev->new_chunk ||
2095 mddev->layout != mddev->new_layout ||
2096 mddev->level != mddev->new_level) {
2097 mddev->new_chunk = mddev->chunk_size;
2098 mddev->new_layout = mddev->layout;
2099 mddev->new_level = mddev->level;
2100 return -EINVAL;
2101 }
2102
2103 md_allow_write(mddev);
2104
2105 raid_disks = mddev->raid_disks + mddev->delta_disks;
2106
2107 if (raid_disks < conf->raid_disks) {
2108 cnt=0;
2109 for (d= 0; d < conf->raid_disks; d++)
2110 if (conf->mirrors[d].rdev)
2111 cnt++;
2112 if (cnt > raid_disks)
2113 return -EBUSY;
2114 }
2115
2116 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2117 if (!newpoolinfo)
2118 return -ENOMEM;
2119 newpoolinfo->mddev = mddev;
2120 newpoolinfo->raid_disks = raid_disks;
2121
2122 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2123 r1bio_pool_free, newpoolinfo);
2124 if (!newpool) {
2125 kfree(newpoolinfo);
2126 return -ENOMEM;
2127 }
2128 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
2129 if (!newmirrors) {
2130 kfree(newpoolinfo);
2131 mempool_destroy(newpool);
2132 return -ENOMEM;
2133 }
2134
2135 raise_barrier(conf);
2136
2137 /* ok, everything is stopped */
2138 oldpool = conf->r1bio_pool;
2139 conf->r1bio_pool = newpool;
2140
2141 for (d = d2 = 0; d < conf->raid_disks; d++) {
2142 mdk_rdev_t *rdev = conf->mirrors[d].rdev;
2143 if (rdev && rdev->raid_disk != d2) {
2144 char nm[20];
2145 sprintf(nm, "rd%d", rdev->raid_disk);
2146 sysfs_remove_link(&mddev->kobj, nm);
2147 rdev->raid_disk = d2;
2148 sprintf(nm, "rd%d", rdev->raid_disk);
2149 sysfs_remove_link(&mddev->kobj, nm);
2150 if (sysfs_create_link(&mddev->kobj,
2151 &rdev->kobj, nm))
2152 printk(KERN_WARNING
2153 "md/raid1: cannot register "
2154 "%s for %s\n",
2155 nm, mdname(mddev));
2156 }
2157 if (rdev)
2158 newmirrors[d2++].rdev = rdev;
2159 }
2160 kfree(conf->mirrors);
2161 conf->mirrors = newmirrors;
2162 kfree(conf->poolinfo);
2163 conf->poolinfo = newpoolinfo;
2164
2165 spin_lock_irqsave(&conf->device_lock, flags);
2166 mddev->degraded += (raid_disks - conf->raid_disks);
2167 spin_unlock_irqrestore(&conf->device_lock, flags);
2168 conf->raid_disks = mddev->raid_disks = raid_disks;
2169 mddev->delta_disks = 0;
2170
2171 conf->last_used = 0; /* just make sure it is in-range */
2172 lower_barrier(conf);
2173
2174 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2175 md_wakeup_thread(mddev->thread);
2176
2177 mempool_destroy(oldpool);
2178 return 0;
2179 }
2180
2181 static void raid1_quiesce(mddev_t *mddev, int state)
2182 {
2183 conf_t *conf = mddev_to_conf(mddev);
2184
2185 switch(state) {
2186 case 1:
2187 raise_barrier(conf);
2188 break;
2189 case 0:
2190 lower_barrier(conf);
2191 break;
2192 }
2193 }
2194
2195
2196 static struct mdk_personality raid1_personality =
2197 {
2198 .name = "raid1",
2199 .level = 1,
2200 .owner = THIS_MODULE,
2201 .make_request = make_request,
2202 .run = run,
2203 .stop = stop,
2204 .status = status,
2205 .error_handler = error,
2206 .hot_add_disk = raid1_add_disk,
2207 .hot_remove_disk= raid1_remove_disk,
2208 .spare_active = raid1_spare_active,
2209 .sync_request = sync_request,
2210 .resize = raid1_resize,
2211 .check_reshape = raid1_reshape,
2212 .quiesce = raid1_quiesce,
2213 };
2214
2215 static int __init raid_init(void)
2216 {
2217 return register_md_personality(&raid1_personality);
2218 }
2219
2220 static void raid_exit(void)
2221 {
2222 unregister_md_personality(&raid1_personality);
2223 }
2224
2225 module_init(raid_init);
2226 module_exit(raid_exit);
2227 MODULE_LICENSE("GPL");
2228 MODULE_ALIAS("md-personality-3"); /* RAID1 */
2229 MODULE_ALIAS("md-raid1");
2230 MODULE_ALIAS("md-level-1");
kernel source for telekom puls (alcatel/mediatek)